As someone who worked in this area:
> Google is also giving health records another whirl. The new initiative, called Care Studio, is aimed at doctors rather than patients. Google’s earlier efforts in this area were derailed in part by hospitals’ sluggishness in digitising their patient records.
Not just digitizing but also in sharing them. Hospital systems are very very reluctant to send out a copy of their records, or even to let you query their system. Especially in the early days of EHRs there was very much a sense that it was their data and they didn't want to let data out because interoperability was a threat to their data moats.
That's good in a privacy sense, but it really did slow down adoption of interoperability initiatives. Finally Medicare/Medicaid basically forced it iirc... if you want Medicare dollars you will send Medicare your encounter data. So now you have to code your records at least good enough to at least get your medicare reimbursements.
> That problem has mostly gone away but another has emerged, says Karen deSalvo, Google’s health chief—the inability of different providers’ records to talk to each other. Dr de Salvo has been vocal about the need for greater interoperability since her days in the Obama administration, where she was in charge of co-ordinating American health information technology. Until that happens, Care Studio is meant to act as both translator and repository (which is, naturally, searchable).
Dead on. It's a hard hard problem though because during the early days we went through a Cambrian explosion in data diversity. Data is often coded very inconsistently or in ways that make it "invalid" to modern schemas. The "miracle of HL7" is basically just a fancy text format, and like the on-table records or XML documents it's still possible to code things in ways that still won't be parseable even if they're syntactially/lexically correct.
There are standardized schemas (f.ex NEDSS schema) available, but they are super over-engineered for a lot of use-cases and can perform quite poorly due to excessive "anything can map to anything" design and field-table-value formats. And again, they suffer from the "there's multiple ways you could code this" thing, just using NEDSS doesn't mean you interoperate with anyone, not close. Not sure what EHRs typically have underneath but it wouldn't surprise me if they all had their own solutions too, EHR software is universally dogshit.
Handling poorly-validated/normalized data from tens of thousands of facilities all doing their own things is a very tough problem, and honestly the only solution is to decentralize it a little bit, facilities have to look at their own data and see if it makes sense the way they're sending it or even if they're storing it wrong. Once it's just another record in a database, if it's not consistent and accurate it likely never will be. It'll just be yet another System-Specific-Record-Identifier in the list for that record, and since it's not coded right nobody will ever be able to read it unless it's patched by someone/some layer.
Data clearinghouses are uniquely positioned to gatekeep that a little bit, and over time there are tools getting written to help various facilities validate their data and see whether it's semantically reasonable. Almost everybody has errors especially if this is their first integration, but they can be brought up to speed on their ongoing record submissions, at least.
I'm sure google would love to be in charge of everyone's health data, or at least to get to see it. Commercial activity isn't notionally allowed but I wonder if HIPAA applies to aggregated data like that, they can probably set up The Business-Purpose Side with a firewall from the Gets To Make Money Querying Anonymized Data side.
It's true that this is a thing where there are intellectual economies-of-scale though. If you can get to One Standard To Rule Them All and actually make it work and not suck, that's useful, everyone just agrees to adopt Google Validation Rules and a shit-ton of terrible mapping/correcting labor gets saved once people get on board. There's just also not really anyone trustworthy enough to handle being the One Giant Repository Of Health Data, Google owning it eventually ends up in massive data mining of sensitive PHI.